EUVIMEDOpdag EUVIMED
Open access · CC BY 4.0

Semi-supervised multimodal survival learning for esophageal cancer using asymmetric feature fusion

Xinpei Zhao, Zelin Wu, Weili Wang, Ying Qiao, Fenglian Li

Array · 2026-12 · 32 · 101430

DOI: 10.1016/j.array.2026.101430 ↗

At a glance

Journal article in Array, 2026-12. Written by Xinpei Zhao and 4 co-authors. 40 references. The full text is freely available (open access, CC BY 4.0).

AbstractBibliografiske oplysningerCiteRelated studies

Abstract (original language)

Survival prediction from computed tomography (CT) and clinical data is challenging when outcome annotations are limited. We propose an uncertainty-aware semi-supervised multimodal survival network (UAS-Net) combining masked CT reconstruction, asymmetric feature fusion, and variance-guided Mean Teacher learning. Imaging and clinical features are projected to 16 and 48 dimensions, respectively, and stochastic teacher predictions guide consistency weighting and sample selection. A retrospective cohort of 187 patients with esophageal cancer was divided into 150 development and 37 held-out test patients. The main configuration used 60 outcome-labeled and 60 outcome-withheld training patients per development fold, with 30 validation patients. UAS-Net achieved a development-validation concordance index (C-index) of 0.7475 ± 0.0283 and a five-model test-ensemble C-index of 0.7493 (95% confidence interval [CI], 0.6239–0.8800), the highest point estimates among the primary comparators. The time-dependent areas under the receiver operating characteristic curve (AUCs) at 1, 2, and 3 years were 0.7329, 0.7616, and 0.7685. The 1–3-year integrated Brier score (IBS) was 0.1835 (95% CI, 0.1375–0.2314). Supplementary sensitivity analyses favored the default variance coefficient and 16:48 projection allocation within the examined settings. These findings demonstrate the potential of semi-supervised CT–clinical integration for survival risk modeling with limited training outcome annotations.

Cite

APA

Zhao, X., Wu, Z., Wang, W., Qiao, Y., & Li, F. (2026). Semi-supervised multimodal survival learning for esophageal cancer using asymmetric feature fusion. Array, 32, 101430. https://doi.org/10.1016/j.array.2026.101430

Vancouver

Zhao X, Wu Z, Wang W, Qiao Y, Li F. Semi-supervised multimodal survival learning for esophageal cancer using asymmetric feature fusion. Array. 2026;32:101430. doi:10.1016/j.array.2026.101430

BibTeX
@article{Zhao2026,
  author = {Zhao, Xinpei and Wu, Zelin and Wang, Weili and Qiao, Ying and Li, Fenglian},
  title = {Semi-supervised multimodal survival learning for esophageal cancer using asymmetric feature fusion},
  journal = {Array},
  year = {2026},
  volume = {32},
  pages = {101430},
  doi = {10.1016/j.array.2026.101430}
}
Start for free →